多源信息驱动的大规模遥感图像场景语义检索

遥感图像检索是对海量空间观测数据进行组织管理和快捷调用的基础，具有重要的理论研究意义和实际应用价值。本项目针对大规模遥感图像场景语义检索问题，从检索问题中的特征提取、度量学习、相关反馈三个方面入手，利用数据驱动和深度学习的方法，研究多源信息监督的遥感图像语义特征学习模型、多模态遥感数据之间的度量学习模型和遥感图像检索中的人机交互相关反馈策略，提出面向遥感数据特性的高层语义检索框架，在降低人工标注代价的同时，实现大规模遥感图像检索效率和性能的提升。本项目的研究有望在大规模遥感图像检索问题上取得显著进展，并为多平台对地观测信息的高效挖掘提供算法和技术支持。

Remote sensing image retrieval is the basis for the organization, management and quick invocation of massive spatial observation data. It has significant theoretical research and practical application value. Our project aims at the semantic retrieval of large-scale remote sensing scenes, starting from three main aspects: feature extraction, metric learning and related feedback problems in retrieval. Based on data-driven and deep learning methods, we will study how to learn high-level features of remote sensing scenes with multi-information supervision, how to learn similarity metric with multi-source remote sensing data, and how to achieve relevant feedback using human-machine interactive information. We intend to propose a high-level semantic retrieval framework for remote sensing data, which can improve the efficiency and performance of large-scale remote sensing scene retrieval and at the same time reduce the labor of manual annotation. Our project is expected to make significant progress in large-scale remote sensing image retrieval and provide algorithms as well as technical support for efficient multi-source earth observation information mining.